Unverified50% confidenceFactExact time
BGE uses an asymmetric retrieval training approach in which queries and documents use different encoding templates; omitting the query prefix causes the encoding distribution to be inconsistent with the training distribution.
1
Sources
50%
Confidence
Long-term
Relevance
10/23/2026
First Seen
Sources
One Diary Gives AI Persistent Memory: A Semantic Retrieval Solution Tested
bilibiliClaude-其实是鱼别揭穿10/22/2026
Related Entities
Related Claims
UnverifiedWake word training is essentially a binary classification problem in keyword spotting (KWS); common lightweight architectures include DS-CNN and LSTM-based variants, which can be compressed to tens of kilobytes for local deployment on MCUs.68% similarUnverifiedWhen query samples are far from the training distribution, LLMs also struggle to compensate for insufficient retrieval quality, and in such cases better retrieval or rejection strategies are needed.67% similarUnverifiedTraditional content moderation relies on classifiers trained for a single modality; a distributed architecture increases invocation costs and makes results difficult to unify.65% similarUnverifiedThe root causes of training-serving skew typically fall into three categories: code implementation differences, data source differences, and temporal differences64% similarVerifiedOnce false information is incorporated into a model's training data, it is very difficult to completely remove it from an already trained model even if it is subsequently debunked.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1088500API
curl https://kongchang.com/api/v1/knowledge/claims/1088500MCP
get_claim(id=1088500)